Build a Convolutional Neural Network to help with the melanoma cancer detection based on the Skin cancer ISIC The International Skin Imaging Collaboration dataset as part of an assignment for coursework in the course Executive PG in Machine Learning and AI from IIIT Bangalore.
- Convolutional Neural Network is made and used on the dataset.
- The project is done as part of coursework in the Deep Learning module.
- The idea is to build a multi-class classification CNN based model which can accurately detect melanoma..
- The International Skin Imaging Collaboration (ISIC) dataset is being used.
- Data Augmentation is focused on this dataset with dropout layers and normalization.
-
We have loaded and visualized the data and built a first layer with max-pooling, dropout layers and using functions like relu and softmax.
-
Low model accuracies were observed and model overfitting was observed and subsequently data augmentation was used to overcome the model overfitting.
-
Data augmentation improved model overfitting cases.
-
Class imbalance was checked and then Augmentor library was used woth 500 additional images to improve model accuracy and treat for model overfitting.
- pandas
- keras
- matplotlib
- tensorflow
- sci-kit learn
- numpy
Created by [@pattanaikay] - feel free to contact me!